[research summary]
Understanding Aggregate Fluctuations: The Importance of Building from Microeconomic
Evidence
John C. Haltiwanger*
* Haltiwanger is an NBER Research Associate in the Programs on Economic
Fluctuations and Growth and Productivity and a Professor of Economics at the
University of Maryland. His “Profile” appears later in this issue.
In recent research using longitudinal establishment-level data, a pervasive
finding is that idiosyncratic factors dominate the distribution of growth rates of output,
employment, investment, and productivity across establishments. Seemingly similar
plants within the same industry exhibit behave quite differently in terms of real activity at
cyclical and longer-run frequencies. Even in the fastest-growing industries, a significant
fraction of establishments decline substantially; similarly, a large fraction of
establishments in the slowest-growing industries grow dramatically. During severe
recessions virtually all industries decline, but within each industry a substantial fraction
of establishments grow. Likewise, during robust recoveries, a substantial fraction of
establishments contract. Simply put, the underlying gross microeconomic changes in
activity dwarf the net changes that we observe in published aggregates.
The tremendous observed within-sector heterogeneity raises a variety of
questions for our understanding and measurement of key macro aggregates. Much of
1
macroeconomic research and our measurement of aggregates is predicated on the
view that building macro aggregates from industry-level data is sufficient for
understanding the behavior of the macro economy. The implicit argument is that, at
least at the detailed industry level, the assumption of a representative firm or
establishment is reasonable.
The finding of tremendous within-industry heterogeneity is not by itself sufficient
to justify abandoning this useful assumption. There is undoubtedly considerable
canceling out of the impact of idiosyncratic shocks (for example, taste, cost, and
technology) that underlie the heterogeneous fortunes across individual producers.
Evidence from recent establishment-level studies of employment, investment, and
productivity growth, however, suggests that this canceling out is far from complete. It is
becoming increasingly apparent that changes in the key macro aggregates at cyclical
and secular frequencies are best understood by tracking the evolution of the crosssectional distribution of activity and changes at the micro level.
A number of different factors are potentially important in this context. The
observed heterogeneity in output, employment, and investment growth rates within
sectors implies a large, continuous pace of reallocation of real activity across production
sites. Such reallocation inherently involves substantial frictions. One obvious and
important friction is that it is time- and resource-consuming for workers (and for other
inputs) to reallocate across production sites. High- and low-frequency changes in key
macro aggregates are likely associated with the interaction of these frictions and the
pace of reallocation. The level of unemployment, as well as the growth rate of
aggregate measures of real activity (for example, real output or productivity), will reflect
2
the efficiency of the economy in accommodating the pace of reallocation. Changes in
institutions, regulation, the pace of technological change, and the sectoral mix of activity
all may alter the intensity of reallocative activity and the economy’s ability to
accommodate the reallocation.
Relatedly, it is important to consider the nature of the adjustment costs at
individual production sites in changing the scale and scope of activity. Accumulating
evidence of lumpy microeconomic adjustment of inputs such as employment and
capital suggests the presence of nonconvexities in micro-level adjustment costs, or, at a
minimum, it implies highly nonlinear adjustment at the micro level. The combination of
nonlinear micro adjustment with micro heterogeneity has important implications for
aggregate fluctuations. One key implication is time-varying elasticities of aggregates
with respect to aggregate shocks. Roughly speaking, time-varying elasticities arise in
this context because the impact of an aggregate shock depends on the distribution of
individual producers’ relative positions to their adjustment thresholds. From this
perspective, characterizing aggregate fluctuations requires tracking how the distribution
of shocks and adjustments has evolved.
Job Creation and Destruction
Much of the recent empirical analysis documenting and analyzing the connection
between micro heterogeneity and aggregate fluctuations has focused on employment
dynamics. My recent work, much of it with Steven J. Davis, focuses on job creation and
destruction.1 Job creation is defined as the sum of employment gains at expanding and
3
new establishments. Job destruction is defined as the sum of employment losses at
contracting and closing establishments. In manufacturing (the sector with the most
readily available establishment-level data for the longest period), annual job creation
and destruction rates are large in absolute terms. In a typical year, roughly one in ten
manufacturing jobs is created and one in ten jobs is destroyed. In nonmanufacturing
(with spottier information based on tabulations from selected states for relatively short
sample periods), job creation and job destruction rates are slightly higher on average.
The large pace of implied job reallocation (measured as the sum of job creation
and job destruction) in both manufacturing and nonmanufacturing sectors highlights the
remarkable fluidity in the distribution of job opportunities across locations in the U.S.
economy. Much of this fluidity reflects shifts within narrowly defined sectors, rather than
between sectors. For example, only 13 percent of job reallocation in manufacturing
reflects shifts of employment opportunities between four-digit sectors.
One important issue for the relevance of these statistics for aggregate
fluctuations is the nature of time-series variation in the pace of job reallocation. In U.S.
manufacturing, the pace of job reallocation varies systematically throughout the cycle at
annual and quarterly frequencies. During downturns, job destruction rises sharply and
job creation falls relatively mildly. Given the observed magnitude and time-series
variation of job reallocation, even modest frictions are likely to yield important
implications for aggregate fluctuations. In recent years, some economists have begun
developing theories to explain the magnitude and cyclical behavior of job (and worker)
flows and the connection to aggregate fluctuations.2 Two types of theories have
received the most attention. One treats fluctuations over time in the intensity of
4
allocative shocks as an important driving force behind aggregate fluctuations. The other
maintains that aggregate shocks are the primary driving forces underlying business
cycles, but that the propagation of aggregate shocks involves intertemporal substitution
effects changing the incentives for the timing of reallocation. Of course, there is an
important debate about the direction of causality and thus the relative contribution of
aggregate and allocative disturbances to aggregate fluctuations. 3 Regardless of the
direction of causality, though, the relevant point is that understanding aggregate
fluctuations requires tracking how the distribution of microeconomic changes has
evolved.
Nonlinear Micro Adjustment
Thus far I have focused on the aggregate consequences generated by the
resource- and time-consuming nature of reallocation. A closely related issue is that the
adjustment at the individual producer level may be nonlinear. For example, about twothirds of annual job creation and destruction are accounted for by establishments with
growth rates above 25 percent in absolute magnitude. Of this group, plant start-ups
account for 12 percent of annual job creation, while plant shutdowns account for about
23 percent of annual job destruction. Thus the distribution of establishment-level
employment changes exhibits both considerable heterogeneity and fat tails. The lumpy
changes at the micro level in combination with the heterogeneity in turn have
consequences beyond those discussed earlier.
Building on the literature about the aggregation of (S,s) models, a useful means
5
of organizing micro data to characterize the interaction of nonlinear micro adjustment
and heterogeneity is the adjustment hazard framework. My work with Ricardo J.
Caballero and Eduardo M. Engel has used this approach to characterize the micro and
macro employment dynamics.4 Using a measure of the gap between desired and actual
employment at the micro level, the adjustment hazard measures the relationship
between the size of this gap and the fraction of it that is closed by the establishment.
The standard convex adjustment cost model implies a constant (flat) hazard, but our
findings using micro data reveal a highly nonlinear hazard, with businesses with large
absolute gaps closing a disproportionately high fraction of the gap. The combination of
a nonlinear micro hazard and considerable micro heterogeneity in the cross-sectional
distribution of the gaps has important implications for aggregate adjustment. Timevarying aggregate elasticities of aggregate employment emerge as the impact of an
aggregate shock depends on the underlying cross-sectional distribution at the time of
the shock and the endogenous dynamics of the cross-sectional distribution interacting
with the nonlinear micro adjustment. Our findings indicate that the marginal
responsiveness for employment varies as much as 70 percent over time. Furthermore,
the impact of the time-varying marginal response is especially large during recessions;
for example, the decline in the 1974–5 recession was 59 percent larger than it would
have been in the absence of nonlinear adjustment.
Investment Dynamics
Nonlinearities in the adjustment dynamics of capital, driven by irreversibilities and
6
related nonconvexities in the adjustment costs of capital, have analogous implications
for aggregate investment dynamics. Several recent studies of establishment-level
investment dynamics support the view that micro investment dynamics exhibit lumpy
adjustment. Plant-level investment is dominated by large-scale investment episodes.
Denoting these large-scale investment episodes as spikes, Russell Cooper, Laura
Power, and I show that the probability of an investment spike is increasing in the time
since the previous spike, lending additional support to the view of a microeconomic
environment with nonconvexities in the adjustment technology.5 Using the adjustment
hazard approach in this context, my work with Caballero and Engel shows a highly
nonlinear relationship between investment and fundamentals. 6 For plants with positive
excess capital, the adjustment hazard is quite flat and close to zero, which is consistent
with irreversibilities in investment. In contrast, plants with large shortages of capital
adjust proportionally more than do plants with small shortages of capital.
As with employment dynamics, the nonlinear adjustment hazard yields timevarying elasticities of aggregate investment with respect to aggregate shocks. For
investment, the marginal responsiveness is highly procyclical and varies by as much as
70 percent. The time-varying elasticities suggest a possible explanation for the oftenpuzzling response of aggregate investment to cost of capital and other shocks. The
basic idea is that the empirical aggregate investment literature has difficulty in
quantifying the relationship between aggregate investment and the cost of capital
because of the failure to incorporate the time-varying responsiveness generated by the
interaction of nonlinear micro adjustment and heterogeneity.
7
Productivity Dynamics
Several of the findings discussed earlier raise a variety of conceptual and
measurement questions regarding our understanding of aggregate productivity growth.
Several key, related findings are of interest. First, there is large-scale, ongoing
reallocation of outputs and inputs across individual producers. Second, the pace of this
reallocation varies over time (both secularly and cyclically) and across sectors. Third,
much of this reallocation reflects within-sector rather than between-sector reallocation.
In addition, recent evidence shows large differentials in the levels and rates of
productivity growth across establishments within the same sector. The rapid pace of
output and input reallocation along with differences in productivity levels and growth
rates are necessary for the pace of reallocation to play an important role in aggregate
(that is, industry) productivity growth. My recent work with Lucia Foster and C. J. Krizan
suggests that reallocation plays a significant role in the changes in productivity growth
at the industry level.7 While measurement-error problems cloud the results somewhat,
two aspects of the results clearly point in this direction. First, our results show a large
contribution from the replacement of less productive exiting plants with more productive
entering plants when productivity changes are measured over five- or ten-year horizons.
Second, the contribution of net entry is disproportionate — that is, the contribution of
net entry to productivity growth exceeds that which would be predicted by simply
examining the share of activity accounted for by entering and exiting plants. These
results are particularly striking for selected service-sector industries that we investigate.
There is tremendous reallocation of activity across service establishments, with much of
8
this reallocation generated by entry and exit. The productivity growth in the selected
service industries we examine is dominated by entry and exit effects. For example, the
primary source of productivity growth between 1987 and 1992 for the automobile repair
shop industry is accounted for by the exit of very low productivity plants.
Endnotes
1
For an overview of this work, see S. J. Davis and J. C. Haltiwanger, “Gross Job
Flows,” in Handbook of Labor Economics, O. Ashenfelter and D. Card, eds..
Amsterdam: North Holland, forthcoming; and S. J. Davis, J. C. Haltiwanger, and S.
Schuh, Job Creation and Destruction, Cambridge: MIT Press, 1996.
2
See, for example, R. J. Caballero and M. Hammour, “On the Timing and
Efficiency of Creative Destruction,” NBER Working Paper No. 4768, June 1994;
published in Quarterly Journal of Economics, 111 (August 1996), pp. 805–52; and D.
Mortensen and C. Pissarides, “New Developments in Models of Search in the Labor
Market,” in Handbook of Labor Economics, O. Ashenfelter and D. Card, eds.
Amsterdam: North Holland, forthcoming.
3
See, for example, S. J. Davis and J. C. Haltiwanger, “Driving Forces and
Employment Fluctuations: New Evidence and Alternative Explanations,” NBER Working
Paper No. 5775, September 1996.
4
R. J. Caballero, E. M. Engel, and J. C. Haltiwanger, “Aggregate Employment
9
Dynamics: Building from Microeconomic Evidence,” NBER Working Paper No. 5042,
February 1995; published in American Economic Review, 87 (March 1997), pp. 115–37.
5
R. Cooper, J. C. Haltiwanger, and L. Power, “Machine Replacement and the
Business Cycle: Lumps and Bumps,” NBER Working Paper No. 5260, September
1995; forthcoming in American Economic Review.
6
R. J. Caballero, E. M. Engel, and J. C. Haltiwanger, “Plant-Level Adjustment
and Aggregate Investment Dynamics,” Brookings Papers on Economic Activity, 2
(1995), pp. 1–39.
7
L. Foster, J. C. Haltiwanger, and C. J. Krizan, “Aggregate Productivity Growth:
Lessons from Microeconomic Evidence,” NBER Working Paper No. 6803, November
1998.
10